OpenAI's ChatGPT Work does the whole job. Own the workflow.

OpenAI shipped ChatGPT Work today, and the pitch is not "a better chatbot." It's an agent that takes an outcome, gathers what it needs from your connected apps and files, breaks the job into steps, and works on it for hours — returning a finished spreadsheet, deck, doc, or web app. It answers less and produces more. For anyone running a small business on a stack of manual processes, that's either the best hire you'll make this year or a quiet way to hand your operations to a vendor. Both are true at once, which is exactly why it's worth thinking about before you point it at real work.

What actually happened

Per OpenAI's own announcement, ChatGPT Work is an agent inside ChatGPT that "can gather information across your apps and workflows to create finished materials like sheets, slides, docs, and web apps, and stay with complex projects for hours by breaking them into smaller steps." It's powered by GPT-5.6, which rolled out the same day, and has OpenAI's Codex coding technology built in — the piece that lets it move from answering questions to actually assembling output.

Two features tell you where this is going. Scheduled Tasks lets ChatGPT run on its own — OpenAI's example is turning new Microsoft Teams and Slack messages into updated docs or slides and pushing the changes to your team, unattended. And Sites in ChatGPT (public beta) turns a request into a live, shareable site or web app behind a URL. It's available on every plan on desktop today, rolling out to Plus, Pro, Business, Enterprise, and Edu on web and mobile over the next few days, per OpenAI. This is the same product category as Claude's background agents — the whole industry is racing to sell you an agent that works while you don't.

Why it matters for your business

Here's the operator's angle, and it's not "don't use it." Use it. It's genuinely good at the grunt work — reconciliations, recurring reports, first-draft decks. The trap is what you feed it. To make ChatGPT Work useful, you teach it your templates, your naming conventions, your approval steps, the logic of how your business actually runs. Do that inside ChatGPT's agent builder and you've quietly moved your operating procedures into a system you don't own and can't export — priced, throttled, and versioned by OpenAI.

The fix is a boundary. Keep the workflow logic — the rules, the steps, the source-of-truth data — in systems you control, and treat ChatGPT Work as one execution surface you point at it, not the place your process lives. That way the day a cheaper or faster agent shows up (and on this release cadence, it will), you swap the executor without rebuilding your operations from memory. Rent the labor. Own the workflow.

Sources: OpenAI — ChatGPT is now a partner for your most ambitious work, OpenAI — ChatGPT Work with GPT-5.6.